More macro evidence of hard landing for US home prices

The US edition of the weekend Financial Times ran the following front-page headline:  Plunge in sales of new homes stokes fear of a hard landing.  The on-line edition headlines Signs of slowdown in US housing.  Sales of new US homes plunged 10.5 per cent last month, prices fell and the stock of unsold homes hit its highest level in 10 years, providing the clearest indications yet that the red-hot housing market may be cooling.

A big slowdown could prompt consumers to cut spending and boost savings. Such a shift could help reduce the current account deficit but at the cost of significantly slower growth.

The consensus of unbiased economists and analysts is largely that the US housing market is poised for a slowdown/correction.  As of yet, however, there is little agreement as to whether the correction will be a "soft landing" or a sharper crash.  In fact, most financial media has tended to emphasize reasoning for a soft landing.  That is, until recently.  Over the past couple of months articles warning of a sharp correction in US real estate have migrated from obscure columns buried deep inside to large, front-page headlines.

More interesting quotes from this weekend's FT:

• New home sales slid to 1.08m, the fourth consecutive fall.  The drop was led by the West Coast, where sales dived 29 per cent, while the price of new homes ... fell 2.9 per cent from a year ago.
• With a flood of new properties on the market, at the present pace of sales, it will take 6.3 months to clear the backlog.
• New home sales are softening fast ... Affordability has declined to a near 15-year low.
• ...economists stressed that the slowdown in February was focused on the West Coast...
• Ian Morris, economist at HSBC, said he believed a hard landing for the housing market over the next 12 months was increasingly likely.  "It would not be a surprise if the West Coast were the first market to go.
• The West Coast has the most richly valued property in the US.
• The median house price in California is 9 times median earnings; the average for the past 20 years was 5 times. ... The next most highly valued area is Washington DC at 7.9 times earnings.
• Many buyers have been forced to take on more exotic mortgage products, such as interest-only or negative amortization loans.
• The rising stock of unsold homes could lower prices still further.  Research at HSBC suggests seven month's supply of unsold housing will mean a soft landing for the market.  But if the figure rises towards nine months it will be a hard landing.

The purpose of the discussion here on Capitalism 2.0 is to explore the macroeconomic, micro economic, and financial aspects of what a hard or soft landing means.  Also, what kind of timing models might be appropriate aside from the HSBC inventory driven model?  Any discussion of theory or models is appropriate here (remember, this forum is moderated). 

General discussions can be directed to the TrackBack of this post over at patrick.net.

 

--By Randolph Harrison

References:

• HSBC Global Research:  Macro US Economics January 2006.  A Froth-Finding Mission; Detecting US Housing Bubbles
• New Residential Sales in February 2006 (PDF)
• New Residential Sales in Feburary 2006 (XLS)
• Historical:  New One Family Houses Sold (XLS)
• Historical:  New One Family Houses for Sale (XLS)
• Historical:  Median and Average Sales Price of New Houses Sold (XLS)
• Historical:  New Houses Sold and For Sale by Construction Stage (XLS)
• Historical:  New Houses Sold by Sales Price (1999-2005) (XLS)
• Historical:  New Houses Sold by Sales Price (XLS)
• Historical:  Median and Average Sales Price of New Houses Sold by Region (XLS)
• Historical:  New Houses Sold by Type of Financing (XLS)
• Historical:  Median and Average Sales Price of New Houses Sold by Type of Financing (XLS)

Comments

[randolfe]

Why new home sales are an important indicator:

From the NAR's own analyses

Important housing market leading indicators include:

* Existing-Home Sales
* Pending Home Sales Index
* New-Home Sales
* Housing Starts
* Housing Affordability
* Mortgage Rates
* Mortgage Applications

Two of these indicators are related to new construction of homes; two are related to financing of home purchases, one is related to affordability, one is related to pending sales (of any type) and one is related to sales of existing homes. Even the NAR considers new homes to be a pivotal indicator.

About the indicator itself:

New home sales growth rates are very well correlated with the housing market and the consumer economy. A good summary of this can be found here.

By contrast, interest rates/mortgage rates do not correlate well to either home prices or the consumer economy because there have been cycles during which home prices dropped, but rates were low and vice versa.

What economists point to is that abrupt slowdowns in the rate of new home sales correlate very well with consumer shifts from consumption spending to savings behaviors. These shifts tend to show up in reduced current account deficits, but also slow or negative GDP growth rates. The past three consumer recessions were correlated to new home sales drops.

[randolfe]

From investopedia:

An economic indicator that measures sales of newly built homes. Released by the U.S. Department of Commerce's Census Bureau, it includes both quantity and price statistics. It is considered to be a lagging indicator of demand in the market and to affect mortgage rates.

A new home sale is considered to be any deposit or contract signing either in the year the house was built or the year after it was built.

So, this indicator is leading because it actually affects other indicators, such as mortgage rates (assumedly because mortgage lenders themselves use this indicator in setting their rates).

[randolfe]

Importance of 3-month moving averages:

For the new home sales indicator, a three month moving average is necessary to correct for expected volatility of this metric. The reason the -10.5% rate for last month is alarming is because it is the 4th consecutive drop, and the 3-month moving average is also moving decidedly lower. This is not likely an anomaly but a genuine trend.

Expect the critics of this data and/or the real-estate housing bulls to make statements along the lines of "one month is not a trend". But, a trend has emerged -- even if it is not as dire as the -10.5% figure implies.

[randolfe]

A theoretical factor model

I tend to think of the primary residential housing market in terms of a factor-model. It is anchored and constrained by both financial and economic theoretical points, such as theoretical prices, affordability indices, rent-to-mortgage ratios, inflation, and interest rates. But, it is also driven off those linearly predictive anchoring points by psychological, regional, and differential factors.

We know that RE prices are sticky; more so on the downside. We know that "bubble psychology" can and does occur manifesting itself in speculative behaviors. We know that a buyer/consumer's optimistic biases can directly affect their price purchase decisions. But, we also know that financial realities impose threshold constraints -- any individual cannot deviate from the theoretical price prediction by more than a theoretical financial debt capacity.

What I would like to do is consolidate this into a factor-model which explores and produces these predictions in aggregate. I suspect that there would be interesting things to learn near the threshold constraints. That is, could we use such a model to reasonably convince ourselves that we are very close to such a threshold now and make useful predictions about the near-term future?

[Fewlesh]

"A theoretical factor model"

Really tough problem as usual. This market has many factors not present in other housing downturns, so trying to line up sale prices from 1992 or 1987 would be much different. In previous downturns we had external events drive the economy. What's strange about this time is that the RE is the driver, and so we have to watch it deflate itself, and not to external events.

Psychology is such a major factor propping this bubble up, that it may be impossible predict this.
What would be nice is to link a psychology tracker to prices. Maybe by scanning news sources and tracking keyword statistics to try and track a trend. Information propagation in exponential in nature, so it should in theory to be possible to track the first upticks in the "psychology" of the market by tracking all news sources in aggregate, and then weighting them by some sort of "bias" metric. The bias metric could be culled from the previous reporting based on previous events. ie: RE news sources can be calibrated based on their reporting of housing crashes/prices in the past.

I know some hedgies are doing this by parsing analists(mispelled on purpose) and corellating the keywords to stock behavior. (Funny thing, as just pure gossip, most of hedgies that trade on analisis say to do exactly opposite of what they recommend)

Fewlesh.

[Bryn Washburn]

Really tough problem as usual. This market has many factors not present in other housing downturns, so trying to line up sale prices from 1992 or 1987 would be much different. In previous downturns we had external events drive the economy. What's strange about this time is that the RE is the driver, and so we have to watch it deflate itself, and not to external events.

Excellent point. I wish I had a dollar for every time I heard a variation on "the last housing bust had nothing to do with any bubble --it was caused by the aerospace cut-backs, post-Cold War recession, L.A. riots, Northridge quake (insert non-repeatable six-sigma event here:_________)", etc., etc.

In some ways I agree with the assertion that it WILL be "different" this time. Different in terms of how much of the California economy is now mostly or completely dependent on housing. I wish I were financially/mathematically savvy enough to create a model to accurately quantify / predict exactly how this will all play out, but sadly that's not my forte. A blind guess would be a fairly nasty recession, at least comparable to the last one in terms of unemployment & duration, but possibly even more severe and longer lasting.

But who knows? Like Fewlesh said, there are so many variables it's hard to predict such macro event, and human psychology is one of the most unpredictable variables of them all.

[randolfe]

Comparing New Home Sales and Existing Home Sales

New home sales and existing home sales are released each month at about the same time. Many comparisons are made between the two series, but before doing any comparisons, one must be aware of some definition differences that affect the timing of the statistics.

The Census Bureau collects new home sales based upon the following definition: "A sale of the new house occurs with the signing of a sales contract or the acceptance of a deposit." The house can be in any stage of construction: not yet started, under construction, or already completed. Typically about 25% of the houses are sold at the time of completion. The remaining 75% are evenly split between those not yet started and those under construction.

Existing home sales data are provided by the National Association of Realtors®. According to them, "the majority of transactions are reported when the sales contract is closed." Most transactions usually involve a mortgage which takes 30-60 days to close. Therefore an existing home sale (closing) most likely involves a sales contract that was signed a month or two prior.

Given the difference in definition, new home sales usually lead existing home sales regarding changes in the residential sales market by a month or two. For example, an existing home sale in January, was probably signed 30 to 45 days earlier which would have been in November or December. This is based on the usual time it takes to obtain and close a mortgage.

Effective with January 2005, the National Association of Realtors created a new monthly series to overcome the lagging effect of the existing home sales definition. This new series is called Pending Home Sales and is based on sales of existing homes where the contract has been signed but the transaction has not been closed, making it roughly equivalent to the new home sales definition. Monthly estimates are expressed as an index where the year 2001 has been set to equal 100.0.

[Fewlesh]

Some morning driving thinking lead to this thought:

Maybe we can take a hint from the guys who simulate the response to bird flu pandemics using monte-carlo methods.

http://www.wired.com/news/technology/medtech/0,69157-0.html

Problem as usual, requires big expensive supercomputers, but maybe a start of the art gaming PC and GPU programming can make a tiny dent for a small town market.

Hard landings could be similiar to the way pandemics spread through hosts. We have information infection, propagation, external factors, etc.

We need to come up with a virtual city, with virutal RE agents, buyers, builders and sellers. Insert external events, and let the simulation run to model how housing prices rise or drop from external factors. We can create virtual models of the demanding wife with the poor husband who wants the big house, to the penny pinching first time buyers looking for a fixer upper with the potential for sweat equity. You simulate the entire financial chain with their jobs, salaries, debt levels. You bribe some credit card companies for detailed financial data on selections of persons to estimate purchasing patterns and costs.

You bring in GIS with property data on an existing city, etc.

Actually, this thing could be extended to predict behavior to all sorts of other financial decisions, but I think RE is a good place to start.

Fewlesh.

[Fewlesh]

Sorry, that bird flu link was not working, here's the correct link:

Wired Magazine - Bird Flu

Fewlesh

[TN]

Fewlesh,

I think you would need to do some heavy duty abstraction. The parameter list alone for that model would be daunting. Each object in your model would be a thesis unto itself.

I would use something like a fluid flow rate model. The liquid would be something similar to the money supply, with housing being tiny balloons with various elasticity and rupture characteristics.

[randolfe]

TN,

I agree it would be a daunting parameter problem. But, this would be a perfect application of GA's. I do worry that it would very quickly become a masterwork of psychological modeling. But, I think Fewlesh is recommending more of a predictive model basis than a descriptive model. So, it may not be necessary to fully describe the behaviors of each object; only be able to reasonably predict the relevant behavior (especially as it relates to the relevance of the overall system) of each.

[Fewlesh]

Do not assume this: complex outputs mean complex underlying algorithms or state machines.

Extremely complex behavior can sometimes be the interaction of a population of simple algorithms. (Maybe RE is a case of this. You're going to have to try it to find out)

The hard part of decoding any complex system is to understand what those damn simple algorithms are. It is impossible to decode the simple algorithms from the output complex end state.

Your "fluid flow rate model" is just random walk brownian motion and could easily be modeled by a trivial monte-carlo simulation of each of the actors. (How did I know that? eell, I did lots of simulation of fluid problems using random walks, and the emergent behavior of fluid flow popped out -- also because my professor told me ...)

Start with a foundation design approach. Start with the simplest of models for each of the actors. Add a single layer of complexity and see if it holds and see what happens. If this buys you your fluid rate model with popping balloons, and you get stuck, it was no worse off then the high level "simple" mathmateical PDF. Many times, you'll get lucky, and some unseen pattern will emerge from some simple addition of what you think was an unimportant layer.

While each object could be a thesis, maybe we just need to capture the most basic "algorithms" that capture behaviors we see.

Lets try and figure out what the most basic actors, and algorithms are (this from my armchair economist position in RE), and see if we can't get brownian motion:

RE selling agent
MLS service, newspaper
RE buying agent
owner buyer, investor buyer
owner seller, builder seller, investor seller
mortgage broker
mortgage bank

Does the above list cover the majority of actors involved in the average housing transaction from sell->buy?

[randolfe]

I think you have the relevant actors covered, although there is actor vs. role classification ambiguity when talking about the different flavors of buyers and sellers. I would tend to break down the actors as the objects (BuyerOccupier, BuyerInvestor, etc.), then implement generalizations of behaviors as interfaces of Buyer and Seller.

This way we could try to uncover simple driving autonoma related to actors and roles separately, and observe the result of different combinations.

[Fewlesh]

:Brainstorm on RE simulation:

Lets add the state information to our actors:
First we need assets:

Each actor has cash (I think this defines an actor: entity that can spend/receive cash, which means I would move MLS out of the actor grouping)

second asset class:
property

Third:
Mortgages owned by bank

Ok simplifying actor classes (we forgot appraisers, anyone else in the basic RE food chain?):
Agents, Buyers, Sellers, Appraisers, Brokers, Banks

Asset class for everyone: Cash
Asset class for Buyers/Sellers: Property
Asset class for Banks: Mortgages

Asset classes are swaped using actors

Ok, we have some basic asset state information. Next is working on basic actor algorithms to produce a market.

[TN]

I think you need an additional piece of state information for Buyers/Sellers/Banks: Risk propensity/tolerance. This would a natural parameter in deciding whether events(aka transactions) would be triggered.

[Robert Cote]

Does the above list cover the majority of actors involved in the average housing transaction from sell->buy?

Secondary market buyers and traders.
Commodity Suppliers. (lumber, land)
Labor Suppliers. (incl movers, staging, repairmen)

[TN]

Designing the Playpen:
Create a grid for spatial representation and simplification to use for buyers location constraints.

Environmental Parameters:
Interest Rates
Wage Growth

The seller actor pricing for now could be Sp(t) = sqrt(t). A slow trend downward as long as it doesn't sell. It might a good enough simple model to represent the downward stickiness of the sellers pysche with its exact shape dependent on its various state information.

The buyers pricing function is a little trickier. Have to think about it.

The banks willingness to loan should be straight foward.

[TN]

Oops the pricing function should be Sp(t) = K - sqrt(t).

[randolfe]

I can agree with the pricing function except that there needs to be the addition of exogenous (to the Actor) factors. I am thinking of situations where macro or micro events cause the Actor to sell at any price attainable, in which case t becomes the sensitive variable, not p. Something like this would be necessary to allow for the system to exhibit threshold event behaviors.

Could we use an existing virtual-world environment like Second Life to handle the spacial model, or are we thinking about something purely abstract?

[Fewlesh]

Randolfe:
Much like chemistry, lets try and keep the spatial variables out at first.

Lets try and get the rate constants for the different "chemical" reactions right, and look at aggregate behavior before you start introducing a lot of spatial complexity (aka biology).

I want to see really basic "law of supply and demand" type behavior before I get into a lot of detail.

TK:
I think these fit under sellers:
Commodity Suppliers
Labor Suppliers

They are part of the Seller:Builder model, and can be simply represented by a profit margin or some such algorithm.

"Buyers/Sellers/Banks: Risk propensity/tolerance".
Risk goes into the behavior algorithms that the actors run. It is probably the most important one, but lets get the basic algorithms of the actors down before we add risk/reward/intelligence

Ok lets start with a very very simple notion of value. Assume property assets have a "fixed" inflation free known intrisic price based on square footage, and some arbitrary distribution function.

Ok, a big factor we have to model is time.

Ok, lets start setting down a few state variables really basic algorithms for actors, and leave out all brokers, just the 3 biggies Buyers, Sellers and Banks:

Buyer:
Income -- monthly cash flow
Cash in bank
Time Window
Algorithm:
Maximize house value for dollar paid over all houses(ie, spend all de money!) within time window else buyer disapears (ie he rents, but were not modelling rent right now)

Seller --
Income -- monthly cash flow
Mortgage -- debt/intrest rate
Cash
House (with intrinsic value)
Previous purchase price and date of house
Offered price
Time Window

Algorithm:
Maximize profit (offered price - purchase price) until Time Window expires, then lower offered price by random value or until Bankruptcy

Bank:
FED Intrest rate:
Reserve Rate:
Algorithm:
Analyze Buyers Income to determine
Mortgage size and intrest rates by simple formulas: Maximum debt/income ratio = 33%, minimum rate = FED Funds rate+1%
maximum rate = FED Funds rate+3%

Have a function that maps debt/income in a linear fasion from min-max from debt/income 15%->33%

Bankruptcy Algorithm:
Minimize loss, ie maximize (offered price - loan amount) within sell Time Window (short for a bank), if expire, then lower offered price -20%. If zero, burn house and collect insurance money. (Just kidding)

After these basic algorithms get roughed out, well create population dynamics: which can create buyers, sellers, and houses from known statistical distributions.

By picking really dumb distributions, poisson model for creation of buyers and sellers, and uniform for income/house price, etc, we should be able to get a damn basic demand curve.

Fewlesh

[Fewlesh]

Opps, I forgot how to model intrisic value.

Basic intelligence for buyers and sellers basic on perfect information (unrealistic, but we'll deal with this later)

There is a probability that a buyer buys when offered priceintrisic value.

[Fewlesh]

that got mangles by a filter.

probability buyer will buy increases as offered price/instrinsic value decreases

probability seller will sell increases as
offered price/purchase price increases

(seller only cares about previous purchased price, not intrinsic price)

[randolfe]

Fewlesh,

Good point. Do you have access to Crystal Ball in Excel or recommendations for alternatives which are easy and quick? I can work up something along those lines later this week and push it up here. CB would allow me to embed the distribution functions directly and apply Monte Carlo to them in a simulation.

[Fewlesh]

I'm a CS major with no experience with excel. I'm a big matlab guy, but my current favorite monte-carlo programming language of choice is python (very quick to program, relatively good performance) and analysis-visualization-statstics in matlab.

Python is a free download, and editors are simple and easy.

http://www.python.org/

If you don't have matlab, octave is a good free substitute.

http://www.octave.org/

Another good free one -- probably better than octave is http://www.r-project.org/

I just have zero experience with excel. I feel sorry for you finance guys :)

Fewlesh.

[randolfe]

You're a smart guy. Excel is far better for working out the early econ model than just hacking out code. I'm not a Python guy anyway coming from more of a big-system object design background. If the early conceptual proof proves valuable at all, then we can drop the implied Excel algorithms into code.

[randolfe]

I have Mathematica, but it gives me a headache. I used to have Matlab, but I quit using it when I learned how to do linear programming in Excel.

[Fewlesh]

Mathematica is for the mathematicians. Very powerful when understood and learned well, but matlab is the engineer's hack tool of choice.

"linear programming in Excel", wow, programming in excel, and all I thought it was just some cell spreadsheet thing.

good ole simplex problem:
http://starship.python.net/crew/aaron_watters/pysimplex/

Brings me back to graph theory and algorithms classes. But it's good that you brought up linear programming. Linear constraints may be good model for pretty complex actor behavior latter down the road(nlogn runtime for the n constraints make then efficient too).

I'm going to get some down time for the next week. I'm going to try and get a demand curve from the buyer-seller-bank model when I pop back.

[TN]

Fewlish:

To make the model persistent, there needs to be a mechanism for:

housing construction
buyers turning into sellers
buyers turning into bigger buyers
population growth

Even though you want to keep the models simple at first, the buyer should maximize EXPECTED house value per money spent. This expected value can be generated by the average price of comps over a sliding time window. I think this would give us the upside down parabola we should expect as output.

[Fewlesh]

To give the chemistry example: that's if you let the reaction run for a long time, which is where you want to go -- long term trend analysis, but you're talking about recursion (feedback).

To make it tractable, you should first take an existing initial distribution of states culled from prior knowledge, and make this big assumption: that the actors do not affect these distributions.

We have a known population of sellers (MLS statistics, income statistics, etc), and an estimated population of buyers (http://www.nationmaster.com/country/us/Age_distribution) -- courtesy of Zephyr

Over the long term this is not true, but were trying to get at house price sensitivities over a limited window knowing the current market state and introducing shock events (intrest rate hikes for example): This type of analysis is called perterbation theory (or sensitivity analysis):

http://en.wikipedia.org/wiki/Perturbation_theory

The whole macro simulation of the population dynamics may be a little too big to chew off at first.

Buyer algorithm:
"This expected value can be generated by the average price of comps over a sliding time window"

Think of an integral over each buyer in a jittered window, that have been monte-carlo sampled to buy/not buy

Since P goes up when "value>price", integral(P) over many buyers will be the expectation of maximized price over many sliding windows, or at least I think so.

You can plug in any range of functions that figures out P = f(value,price), it should be monotonic though.

We still haven't created the "marketplace" yet:
In the marketplace groups of buyers/sellers probabilistically compete over the properties through simulation rounds. (Each round is a day for example)

[astrid]

Randy H,

Please moderate the patrick.net thread. All of a sudden it got really ugly.

[randolfe]

Astrid,

Sorry, I wasn't online late last night. If you email me who perpetrated what, I'll take care of it (email link on the side bar).

[randolfe]

"linear programming in Excel", wow, programming in excel, and all I thought it was just some cell spreadsheet thing.

You can actually write entire systems in Excel using VBA. Not the greatest architecture, but you can do some incredible stuff pretty fast (the stuff Excel does well, like 2D array manipulation). I did a classic linear portfolio optimizer (efficient frontier stuff) a while back when i was in grad school. Some folks I know are still using it to balance their 401k.

[astrid]

Randy,

I'll remember that in the future. Thanks for clearing the air on that thread! All in all, it turned out to be insightful because Panicearly took the abuse in good humor and it didn't turn into a flame war.